Polyclonality, Shared Strains, and Convergent Evolution in Chronic Cystic Fibrosis Staphylococcus aureus Airway Infection

Abstract

Rationale: Staphylococcus aureus is the most common respiratory pathogen isolated from patients with cystic fibrosis (CF) in the United States. Although modes of acquisition and genetic adaptation have been described for Pseudomonas aeruginosa, resulting in improved diagnosis and treatment, these features remain more poorly defined for S. aureus.

Objectives: To characterize the molecular epidemiology and genetic adaptation of S. aureus during chronic CF airway infection and in response to antibiotic therapy.

Methods: We performed whole-genome sequencing of 1,382 S. aureus isolates collected longitudinally over a mean 2.2 years from 246 children with CF at five U.S. centers between 2008 and 2017. Results were integrated with clinical and demographic data to characterize bacterial population dynamics and identify common genetic targets of in vivo adaptation.

Measurements and Main Results: Results showed that 45.5% of patients carried multiple, coexisting S. aureus lineages, often having different antibiotic susceptibility profiles. Adaptation during the course of infection commonly occurred in a set of genes related to persistence and antimicrobial resistance. Individual sequence types demonstrated wide geographic distribution, and we identified limited strain-sharing among children linked by common household or clinical exposures. Unlike P. aeruginosa, S. aureus genetic diversity was unconstrained, with an ongoing flow of new genetic elements into the population of isolates from children with CF.

Conclusions: CF airways are frequently coinfected by multiple, genetically distinct S. aureus lineages, indicating that current clinical procedures for sampling isolates and selecting antibiotics are likely inadequate. Strains can be shared by patients in close domestic or clinical contact and can undergo convergent evolution in key persistence and antimicrobial-resistance genes, suggesting novel diagnostic and therapeutic approaches for future study.

At a Glance Commentary

Scientific Knowledge on the Subject

Staphylococcus aureus is an extremely common respiratory pathogen in patients with cystic fibrosis (CF). Nevertheless, the genetic basis of S. aureus adaptation to the CF airway over time, how strains evolve in response to antimicrobial therapy, and whether S. aureus strains are shared between patients with CF remains poorly understood. Lack of knowledge in these key areas presents a barrier to developing improved strategies for preventing and treating S. aureus respiratory infections in people with CF.

What This Study Adds to the Field

In this study, we performed whole-genome sequencing and analysis of 1,382 S. aureus isolates collected longitudinally from 246 pediatric patients with CF from across the United States. We found that people with CF were frequently coinfected by multiple, genetically distinct strains of S. aureus that often differed in antibiotic susceptibility. In some cases, S. aureus strains were shared between patients through mutual home or clinical environments. We identified a common set of S. aureus genes that were frequently mutated as strains adapted over the course of chronic infection, identifying candidate mechanisms contributing to antimicrobial resistance and persistence in the CF airway.

Cystic fibrosis (CF) is characterized by chronic airway infection with a variety of microbes (1). Although Pseudomonas aeruginosa is classically associated with this disease, its prevalence in patients with CF has steadily declined over the past two decades, likely because of an improved understanding of P. aeruginosa pathogenesis (2–6) and resultant improvements in treatment (7, 8). The microbial pathobiology of CF has consequently shifted, such that Staphylococcus aureus is now the most common respiratory pathogen isolated in the United States and other countries (9).

S. aureus infection in CF is especially prevalent among children (9) and is associated with airway inflammation, respiratory exacerbations, and lung-function decline (10). Initial infection often occurs early in life (11) and may persist for years (12) despite aggressive eradication attempts (13). Although CFTR modulator therapies show great promise for improving airway clearance and lung function, studies to date have not found these medications to alter the burden of S. aureus in the CF airway (14, 15). Therefore, a great need remains for improved, evidence-based S. aureus treatments in children with CF (16, 17).

The efficacy of antistaphylococcal therapies is likely limited by adaptive genetic changes occurring in S. aureus during chronic infection, including mutations promoting antibiotic tolerance and persistence in the host (18). For example, we previously found that slow-growing, antibiotic-resistant small-colony variants (SCVs) emerge frequently during chronic CF infections and are more closely associated with lung-function decline than “traditional CF pathogens” (19). Other work has demonstrated an increased capacity of S. aureus to form biofilms as CF infection progresses (20).

Bacterial whole-genome sequencing (WGS) offers opportunities to deeply understand critical features of S. aureus epidemiology, transmissibility, antimicrobial resistance, and adaptation in chronic CF airway infection. However, prior investigations in this area have been limited in scope, geography, and sample size for both patients and bacterial isolates (20–23).

We performed WGS of 1,382 S. aureus respiratory isolates collected longitudinally from 246 children with CF at five geographically dispersed U.S. centers to characterize the molecular epidemiology of S. aureus within and across patients and to identify common mechanisms of bacterial adaptation during infection and in response to antibiotic treatment. These analyses recognize previously undescribed patterns of infection, genetic change, and transmission of S. aureus, suggesting that new approaches to laboratory surveillance, infection control, and treatment are needed for this prevalent CF pathogen.

Methods

Cohort, Sample Collection, and Sequencing

S. aureus isolates were collected from children at five U.S. CF centers (Seattle, Washington; Houston, Texas; Pittsburgh, Pennsylvania; Birmingham, Alabama; and Boston, Massachusetts) between January 2008 and April 2017. Sputum or oropharyngeal swabs were collected at routine study visits and cultured using standard clinical laboratory protocols. Children positive for S. aureus at multiple time points during this period were included. Pulmonary function tests and data on coinfection, antibiotic exposure, and other therapies were recorded at study visits whenever possible. This study was approved by the University of Washington Human Subjects Institutional Review Board (number 51824), and parental informed consent was obtained for each patient.

All S. aureus isolates cultured from included patients, including representative colonies having distinct morphologies, were subjected to WGS using an Illumina NextSeq 500 sequencer (Illumina), as previously described (24).

Analysis of S. aureus Population Structure and Mutation Rates

De novo assembly of sequence reads from each isolate into draft genomes was performed with ABySS version 2.0.2 (25), followed by annotation of gene features with Prokka version 1.12 (26). Analysis of the CF S. aureus “core genome” (essential genetic elements shared by nearly all isolates) and “pangenome” (the collective genetic content of all isolates) was then performed with Roary version 3.11.2 (27). The rate at which new S. aureus genes were observed (n) as a function of total genomes examined (N) was modeled according to Heap’s Law (n ∼ kN^−α) (28, 29). We characterized the overall population structure of isolates and inferred individual mutation rates over time on the basis of phylogenomic reconstruction of core-genome alignments using FastTree version 2.1 (30) and BEAST version 1.10.4 (31), respectively.

Identification of Related Strains across and within Patients with CF

To determine thresholds for defining related strains across and within hosts, we calculated pairwise genetic distances among isolates after identifying sequence variants at each genetic position (“variant calling”) relative to a common reference genome (CP000253.1), as previously described (24). Across all patients, we then modeled incremental thresholds of genetic distance to estimate their test characteristics for correct grouping of strains derived from the same versus another individual. Separately, we defined the limit of genetic variation among isolates related by descent within each individual using an aggregated density plot of pairwise distances among isolates from each subject. After log-transformation, a Gaussian mixture model was used to estimate the bounds of genetic diversity evolved in vivo after a single strain-acquisition event versus arising through multiple acquisition events with unrelated strains.

Analysis of De Novo Mutations and Gene Selection

For each patient-specific “lineage” of isolates determined to be related by descent, we identified the closest S. aureus genome from GenBank (National Center for Biotechnology Information) and performed alignment and variant calling (24). To identify genes commonly under positive selection during adaptation to the CF airway, we cataloged mutations arising between the first- and last-collected isolate from each lineage. The proportion of nonsynonymous versus synonymous mutations relative to the number of isolate pairs lacking mutations in each gene was analyzed using a Fisher exact test with Bonferroni correction.

Additional methodological detail is provided in the online supplement.

Results

Study Population

Two hundred forty-six children from five centers were included. Detailed clinical characteristics were available for 197 (80.2%) subjects (Table 1). An average of 5.6 (range, 2–22) S. aureus isolates were collected per patient over an average of 2.2 years (range, 0.3–6.6), yielding 1,382 total isolates (28.1% methicillin-resistant variants, 15.8% SCVs).

Table 1.

Representative Demographic and Clinical Characteristics of Included Patients

	Birmingham	Boston	Houston	Pittsburgh	Seattle	Overall
Patients, n	38	26	28	23	83	198
Sampling period, yr, mean (SD)	1.9 (0.1)	1.8 (0.2)	1.7 (0.3)	1.8 (0.2)	2.2 (1.3)	2 (0.9)
Isolates per patient, mean (SD)	5.2 (4.3)	3.6 (2.6)	4.2 (4)	5.9 (4)	7.1 (6.2)	5.7 (5.1)
Age at enrollment, yr, mean (SD)	11.7 (2.8)	12.6 (3.1)	12.1 (3.3)	12 (3.1)	9.3 (4)	10.9 (3.7)
Sex, F, n (%)	17 (44.7)	11 (42.3)	20 (71.4)	15 (65.2)	48 (57.8)	111 (56.1)
Mutation type, n (%)
ΔF508 homozygote	20 (52.6)	13 (50)	18 (64.3)	14 (60.9)	51 (62.2)	116 (58.9)
ΔF508 heterozygote	15 (39.5)	13 (50)	10 (35.7)	6 (26.1)	21 (25.6)	65 (33)
Other	3 (7.9)	0 (0)	0 (0)	3 (13)	10 (12.2)	16 (8.1)
Any MRSA isolated during study period, n (%)	27 (71.1)	9 (34.6)	12 (42.9)	16 (69.6)	30 (36.1)	94 (47.5)
Any SCV S. aureus isolated during study period, n (%)	12 (31.6)	4 (15.4)	3 (10.7)	11 (47.8)	27 (32.5)	57 (28.8)
Chronic P. aeruginosa coinfection*, n (%)	7 (18.4)	4 (15.4)	8 (28.6)	3 (13)	30 (36.1)	52 (26.3)
Antibiotic exposure during study period (any), n (%)
Fluoroquinolone	22 (57.9)	10 (38.5)	10 (35.7)	10 (43.5)	26 (31.3)	78 (39.4)
Aminoglycoside	29 (76.3)	9 (34.6)	13 (46.4)	9 (39.1)	35 (42.2)	95 (48)
β-Lactam	18 (47.4)	21 (80.8)	17 (60.7)	19 (82.6)	69 (83.1)	144 (72.7)
Macrolide	33 (86.8)	8 (30.8)	11 (39.3)	8 (34.8)	30 (36.1)	90 (45.5)
Rifamycin	1 (2.6)	0 (0)	1 (3.6)	2 (8.7)	11 (13.3)	15 (7.6)
Tetracycline	10 (26.3)	5 (19.2)	5 (17.9)	2 (8.7)	9 (10.8)	31 (15.7)
TMP–SMX	25 (65.8)	9 (34.6)	9 (32.1)	8 (34.8)	32 (38.6)	83 (41.9)
FEV1% predicted, enrollment baseline, mean (SD)	93.7 (18)	99.6 (17)	95.7 (15.3)	94.2 (12.5)	91.6 (21.2)	94.1 (18.2)
FEV1% predicted, slope over study period (% predicted/yr), mean (SD)	0.66 (4.1)	0.61 (7.5)	−2.7 (4.9)	−5.26 (6)	−0.25 (6.2)	−0.89 (6.3)

Definition of abbreviations: MRSA = methicillin-resistant S. aureus; P. aeruginosa = Pseudomonas aeruginosa; S. aureus = Staphylococcus aureus; SCV = small-colony variant; TMP-SMX = trimethoprim-sulfamethoxazole.

Complete clinical and demographic characteristics were available for 198 of 248 included patients.

^*P. aeruginosa isolated from majority of cultures obtained during study period.

Distribution of Strain Types

WGS allowed confident assignment of 97.7% of S. aureus isolates to an established multilocus sequence type (MLST) (Figure 1, outer ring), the most prevalent of which were type 5 (health care–associated methicillin-resistant S. aureus [MRSA]), type 8 (community-associated MRSA types), and type 30, collectively encompassing 52.4% of typeable isolates. Comparison of MLST assignments with core phylogenomic analysis revealed overlap between certain sequence types (5, 3499, and 105; 45 and 256; 39 and 30; 59 and 87), indicating that they do not represent distinct evolutionary lineages. Methicillin resistance was most prevalent among types 5, 8, 3499, 105, and 474, but many groups included both MRSA and methicillin-sensitive S. aureus (MSSA) isolates, often sharing high degrees of genomic similarity (Figure 1, middle ring). Sequence types were not constrained by geographic location of the source patients, signifying unrestricted dispersal of diverse S. aureus strains across clinical sites (Figure 1, inner ring).

Figure 1.

Phylogenetic reconstruction of Staphylococcus aureus isolate core-genome alignments. The medical center of origin (inner ring), methicillin-resistance phenotype (middle ring), and multilocus sequence type (outer ring) are labeled circumferentially. S. aureus sequence types colonizing the cystic fibrosis airway have broad geographic distribution, with types 5, 8, and 30 predominating. Particular sequence types (5, 8, 3499, and 474) cluster with methicillin resistance, but intermixing of MRSA and MSSA is observed, even within clades of highly related isolates. MRSA = methicillin-resistant S. aureus; MSSA = methicillin-sensitive S. aureus.

Core and Pangenome Content of CF S. Aureus Isolates

In addition to core genes present in all members of a species, bacteria carry a variable number of “accessory” genes present in some isolates but absent from others. To characterize the dynamics of S. aureus gene acquisition and carriage in CF, we analyzed the conserved (core) and variable (accessory) gene content of the isolate cohort (Figure 2). Six hundred fifty-five and 1,142 core genes were present in ≥99% and ≥95% of isolates, respectively, which was compatible with, but slightly lower than, estimates from smaller studies of S. aureus in the general population (28, 32). In the pangenome, 21,358 unique genes were identified, comprising the entire set of genes in the isolate collection.

Figure 2.

Rarefaction curves showing the diversity of gene content among Staphylococcus aureus cystic fibrosis respiratory isolates. Content of the core genome (blue line, gene presence in 99% of strains) can be established with approximately 100 genomes. The size of the pangenome (red line, number of total distinct S. aureus genes observed including both core and accessory genes) continues to expand with the addition of new isolates, reaching a total of 21,358 unique genes in this experiment, and is not predicted to reach an asymptote. When the rate of new gene identification is modeled using Heap’s Law (inset), the pangenome (purple point estimates and fitted line) is seen to lie in an “open” state (nonshaded region), indicating ongoing acquisition of new genes among the total population of S. aureus strains infecting patients with cystic fibrosis.

A species’ pangenome can be characterized as “closed,” if limited to a finite pool of genes, or “open,” meaning that active gene acquisition by members of the species provides unlimited potential for genetic diversity (28, 29). The rate of new genes identified per genome analyzed in this cohort was consistent with an open pangenome (Figure 2), indicating ongoing flow of new genetic elements into S. aureus populations infecting patients with CF.

In Vivo Mutation Rates of S. aureus in CF

Hypermutation has been hypothesized as important feature of S. aureus adaptation and pathogenesis in CF (33). However, in vivo core-genome mutation rates for isolates in this study were estimated at 5.0 × 10⁻⁶ (95% confidence interval [CI], 2.5 × 10⁻⁵ to 1.0 × 10⁻⁶) mutations per nucleotide per year, similar to inferred values for more generalized S. aureus populations (22, 34–36). As expected, rates were significantly elevated (1.0 × 10⁻⁵; P = 2.5 × 10⁻⁵) among the 4.7% of isolates with mutS or mutL nonsense mutations but did not differ between wild-type isolates and SCVs (P = 0.26; Figure 3) or between isolates from patients exposed versus unexposed to any antibiotic class.

Figure 3.

In vivo mutation rates of Staphylococcus aureus isolates in the cystic fibrosis airway are similar to published estimates of S. aureus molecular clock speed in other clinical populations. No difference in mutation rates between wild-type strains and small-colony variants was observed.

Isolate Relatedness and Strain Sharing among Patients with CF

Although epidemiologic studies have identified P. aeruginosa strain transmission among patients with CF (23, 37), less is known about whether similar patterns occur for S. aureus. We calculated optimal thresholds of genetic differences between isolates for defining genetic relatedness and then applied these criteria to ascertain whether patients with CF shared S. aureus lineages having a recent common ancestor. Using predictive receiver operating characteristic (38) and receiver operating characteristic curves (Figures 4A and 4B), we determined that isolate pairs with ∼200 or fewer genomic differences had a high probability of originating from a single patient (positive predictive value [PPV] = 84.3%; negative predictive value [NPV] = 99.8%). By comparison, MLST, which considers only sequence fragments of specific genes, performed poorly in predicting relatedness (PPV = 4.0%; NPV = 99.8%). We identified 42 cases in which isolates separated by ≤200 pairwise genomic differences were shared among two (n = 39) or three (n = 3) different individuals. Patients in each case were geographically clustered, originating from one (n = 30), two (n = 11), or three (n = 1) clinical sites. Shared strains were not enriched for specific MLSTs, methicillin resistance, or SCV status, indicating that these clusters most likely arose from distinct transmission events, rather than from the existence of ubiquitous or specific, highly transmissible strain types.

Figure 4.

(A) Predictive receiver operating characteristic and (B) receiver operating characteristic curves for inter- versus intrahost genetic variation of Staphylococcus aureus isolates, showing test characteristics for correct identification of patient of origin, based on various thresholds of S. aureus pairwise genetic distance. A conservative threshold of 36 or fewer variants had the highest positive predictive value for host identification, whereas a more inclusive threshold of approximately 200 variants offers more balanced test characteristics. Dashed lines indicate corresponding test characteristics for multilocus sequence typing. (C) Gaussian mixture modeling of within-host genetic diversity identified two nonoverlapping distributions of pairwise genetic distance, circumscribing genetic diversity within a lineage of related isolates potentially derived over the lifetime of a host (black) versus highly divergent strains representing acquisition of an unrelated strain (gray). Arrow indicates threshold of 3,000 variants used to distinguish these distributions for the purposes of analysis.

To investigate how the most recent strain-sharing may have occurred, we focused on 11 instances passing a more conservative threshold of ≤36 variants (PPV = 91.8%; NPV = 99.5%). Each event involved a pair of patients from one (n = 7) or two (n = 4) clinical sites, collectively comprising 20 patients. Of the 11 pairs, 5 were siblings, for whom strain-sharing is likely attributable to common household environments. The other six pairs involved unrelated patients. Two pairs involved children from the same center who had at least one clinic encounter on the same day with the same provider preceding identification of the shared S. aureus strain. The remaining four pairs constituted patients from geographically distant regions, three of which spanned the same two centers. Chart review revealed no evidence of relevant patient relocation or transfer of care between centers. However, several involved patients had contact with a single provider who transferred between these sites during the study period; in each case, the strain was first identified at one site and then observed at the other after arrival of the provider in that location. The final cross-center strain-sharing event spanned a different set of centers, with no readily identifiable connection between patients or medical systems.

We conclude that S. aureus strains are occasionally shared between patients with CF, particularly those having close contact. Both community- and health care–associated transmission networks plausibly contribute to these cases.

Multiple Genetically Distinct S. aureus Lineages Colonize the CF Airway

S. aureus genetic diversity within a single patient may arise 1) over time through evolution of a single, founding bacterial clone and/or 2) through coinfection with genetically distinct lineages. To distinguish these possibilities, we performed Gaussian mixture modeling of pairwise genetic distances among all isolates cultured from individual hosts. This analysis demonstrated two distinct distributions of pairwise genetic distance between isolates (Figure 4C) that respectively corresponded to these two sources of within-host diversity. The first had a median of 64 (95% CI, 10–593) variants, consistent with diversity arising from within-host expansion of a single progenitor. The second was centered around ∼39,000 (95% CI, 17,907–54,887) variants, beyond the degree of diversity that could arise within a clonal lineage during a child’s lifetime (see Equation E1 in the online supplement), thereby being most consistent with independent acquisition of multiple, genetically distinct strains. Using a threshold of 3,000 variants to circumscribe these distributions, 112 (45.5%) subjects harbored multiple S. aureus lineages, averaging 1.5 independent lineages (range, 1–4) per patient. To resolve the question of whether such events resulted from concurrent and/or sequential infection, we correlated isolate collection dates with lineage assignments in relevant patients. Most instances were consistent with concurrent (81%), rather than sequential (19%), infection, indicating that simultaneous infection with multiple S. aureus lineages commonly occurs.

Within-Host Adaptation of S. aureus in CF Involves Specific Genes

To identify bacterial genes under selection during chronic CF airway infection, we assessed the ratio of nonsynonymous (conferring an amino acid change) to synonymous (silent) mutations. A significant excess of nonsynonymous mutation occurred in nine genes (Figure 5 and Table 2), signifying positive selection. Three (rpoB, rpsJ, set9) displayed recurrent mutations at specific sites and a low proportion of disruptive (stop-gain–, frameshift-, or large-indel) variants (<10%), consistent with a gain of function (Figures 6A, 6B, and E1). Five other genes (agrA, ebh, fmtB, rsbU, thyA) carried >10% of disruptive mutations distributed throughout the coding sequence, suggesting loss of function (Figures 6C, 6D, and E1).

Figure 5.

Q–Q plot of P values for de novo dN/dS ratios by gene. Each point represents a Staphylococcus aureus gene, and elevation above the dashed line indicates an imbalance in the proportion of protein-altering mutations above what would be expected by chance. Nine genes (red, listed in Table 2) met genome-wide significance for adaptive selection after Bonferroni correction. A horizontal jitter is added to compensate for overplotting and facilitate visualization. dN = nonsynonymous; dS = synonymous; Q–Q = quantile–quantile.

Table 2.

Staphylococcus aureus Genes under Positive Selection during Chronic Cystic Fibrosis Respiratory Infection

Gene	Function	dN/dS Mutation Counts (Ratio)	dN/dS Proportion P Value	Proportion of Disruptive Mutations* (%)
rpoB	DNA-directed RNA polymerase, β subunit	37/3 (12.3)	10−7.2	0.0
fmtB	LPXTG-motif cell wall anchor domain	40/7 (5.7)	10−5.4	11.1
agrA	Accessory gene regulator A	18/0 (∞)	10−5.2	22.2
rpsJ	30S ribosomal protein S10	18/0 (∞)	10−5.2	0.0
rsbU	Sigma factor B regulator protein	18/0 (∞)	10−5.2	11.1
thyA	Thymidylate synthase	21/1 (21)	10−5	42.9
set9	Superantigen-like protein	17/0 (∞)	10−4.9	0.0
walK	Two-component sensor histidine kinase	20/1 (20)	10−4.7	0.0
ebh	Surface protein, ECM binding protein–like protein A	73/26 (2.8)	10−4.5	11.6

Definition of abbreviations: ECM = extracellular matrix; dN = nonsynonymous; dS = synonymous.

^*The proportion of “disruptive” mutations equals the number of stop-gain, frameshift, or large (affecting more than one codon) indels divided by the total number of nonsynonymous mutations observed in the gene.

Figure 6.

Distribution of de novo mutations in four exemplary proteins from Table 2 identified as being under positive selection during chronic Staphylococcus aureus infection in cystic fibrosis. Consensus sequence amino acids are indicated in black; nonsynonymous de novo mutations observed in the sample set are indicated by colored blocks with variant amino acids identified by letters. S. aureus adaptive mutation in some genes, including (A) rpoB and (B) rpsJ, is characterized by missense mutation at specific residues, suggesting gain of gene function in cystic fibrosis. Mutations in other genes such as (C) agrA and (D) thyA are characterized by nonsense mutation throughout the protein, suggesting loss of function. Analogous plots for the five additional genes identified in the nonsynonymous versus synonymous de novo mutation analysis are included in Figure E1. FS = frameshift mutation; ID = non-frameshift insertion/deletion; SG = stop-gain mutation.

Some Genetic Adaptations Are Associated with Antimicrobial Therapy

Several positively selected genes (thyA, rpoB, rpsJ, fmtB, and walK) have known or suspected roles in antibiotic resistance. We tested the association of mutations in these genes with antimicrobial exposures during the study period (Table 1) and compared in vitro antibiotic susceptibilities between mutants and their unmutated progenitors. Nonsense mutations in thyA are known to confer trimethoprim–sulfamethoxazole resistance (39); accordingly, those mutations were strongly associated with trimethoprim–sulfamethoxazole exposure (odds ratio [OR], 1.8; 95% CI, 1.39–2.50; P < 0.0001). Mutations in rpsJ empirically reduced susceptibility to tetracyclines and levofloxacin (Table E2) and were independently associated with tetracycline (OR, 2.09; 95% CI, 1.31–3.55; P = 0.003) and fluoroquinolone (OR, 2.04; 95% CI, 1.29–3.32; P = 0.003) therapy. Similarly, rpoB mutations reduced susceptibility to rifampicin (Table E3) and were associated with rifamycin exposure (OR, 2.93; 95% CI, 1.44–8.24; P = 0.01).

These results indicate that antimicrobial resistance arises in vivo among colonizing S. aureus strains in response to CF treatment and is a prominent but not exclusive factor in selection for such mutations in resistance-associated genes. Whereas most patients harboring de novo thyA (90.1%) and rpsJ (66.7%) mutations had recorded exposure to the corresponding antimicrobials, only 17.4% of patients with de novo mutations in rpoB were exposed to a rifamycin. rpoB mutations conferred no measurable resistance to other CF-relevant antibiotics, arguing against cross-resistance effects and indicating that antibiotic exposure incompletely accounts for rpoB-mutant selection. No association between antimicrobial use or resistance was observed for fmtB or walK mutations (Tables E4 and E5), suggesting that their selection was not driven by antibiotics.

Mutation of Transcriptional Regulators Modulates S. aureus Virulence Factors

Two positively selected genes, agrA and rsbU, encode transcriptional regulators. To investigate the consequences of those mutations, we compared phenotypic changes in hemolysis, pigment production, and protease activity between agrA and rsbU mutants and their unmutated progenitors. Mutations in clinical isolates demonstrated alterations in virulence-factor expression concordant with transposon mutants (Tables E7 and E8), suggesting selection for virulence-modulating, loss-of-regulatory-function mutation in these pathways during chronic CF airway infection.

Association of Gene Selection with Lung-Function Decline

In an exploratory analysis, we separately correlated the degree of positive selection in all S. aureus genes with lung-function change over the 2.2-year average period of study participation. No genes met a genome-wide threshold of significance for differential selection between groups of patients having negative versus positive slopes of FEV₁ percent–predicted change. Below this threshold, citB (aconitate hydratase) shows the greatest difference between groups (P = 0.047) (Table E6) and encodes a tricarboxylic acid cycle enzyme implicated in survival of CF-adapted S. aureus isolates (40) and virulence regulation (41). This hypothesis-generating finding requires replication in a larger, independent cohort.

Methicillin Resistance Is Dynamic within S. aureus Lineages

Eighty-nine of 383 S. aureus lineages contained at least one methicillin-resistant isolate. Unexpectedly, 17 (19.1%) comprised a combination of MRSA and MSSA isolates that were related by descent. The cassette carrying the MRSA-defining mecA gene (SCCmec) exists in numerous subtypes (42) that were consistent across MRSA isolates within each lineage. Most MSSA isolates completely lacked the cassette, except for one carrying a truncated copy of the mecA gene. We identified two primary alleles of the SCCmec insertion site, orfX, one of which was more common in lineages comprising both MSSA and MRSA (46/54 = 85.2%) than those having a stable MSSA genotype (82/248 = 33.1%; P < 0.001) (Figure E3). These observations indicate that genetic rearrangements involving SCCmec occur with measurable frequency within a patient, may be predisposed by polymorphisms near the site of cassette integration, and result in dynamic clinical determinations of patient MRSA status.

Discussion

This large, multicenter study represents the largest WGS analysis of S. aureus to date in CF, or in any single disease group, providing unprecedented insight into bacterial population dynamics and genetic adaptation during chronic respiratory infection. Many of our findings were surprising and have important clinical implications for the diagnosis and treatment of S. aureus infections in CF.

At a population level, S. aureus strains among children with CF in the United States were genetically diverse and achieved broad geographic distribution. Consistent with this high degree of diversity, the pangenome of the population was predicted to be open to gene acquisition, indicating that genetic content of strains infecting children with CF is not restricted (28). This finding agrees with modeling of more general S. aureus collections (28, 32) and other CF-relevant species including Burkholderia cepacia complex (43), Stenotrophomonas maltophilia (44), and Streptococcus pneumoniae (45) but contrasts with the closed pangenomes of P. aeruginosa (46) and S. lugdunensis (47). It is possible this ability to adapt to diverse environments through new gene acquisition has facilitated the increasing prevalence of S. aureus in CF relative to P. aeruginosa (9), which has a larger and well-adapted, but static, genomic repertoire. This genetic flexibility may increase the organism’s metabolic versatility, virulence properties, or survival strategies amid the nutritional and inhibitory challenges encountered in the CF lung, thus complicating how we understand its pathogenic potential and response to antistaphylococcal agents. The clinical implications of this finding require further study.

Because of its propensity for evolution within patients, identifying S. aureus transmission events by genetic similarity is challenging (35), and proposed thresholds for defining clonal-isolate relationships have been inconsistent and largely theoretical, ranging from 10 to 71 variants (48, 49). Using a statistical framework, we empirically determined cutoffs for circumscribing S. aureus isolates related by descent. Using even conservative criteria, we observed recent strain-sharing among at least 20 patients. All but one event could be explained by a common household environment, repeated contemporaneous exposures in the same clinic, or, surprisingly, exposures across distant healthcare centers via a shared provider. Although infrequent, these instances highlight the potential for domestic transmission and the ongoing importance of infection control in outpatient settings (37), including the potential use of prospective, sequencing-based surveillance.

Discrimination between MSSA and MRSA is important to current CF clinical practice because it is used to inform antibiotic treatment (50) and because MRSA infection predicts worse disease outcomes (51). We found many patients were concurrently infected with multiple, genetically distinct S. aureus lineages, often differing in antibiotic susceptibility. This observation is contrary to the concept that patients with CF are typically colonized with a single, founding S. aureus clone that persists over time (12), with mixed-strain S. aureus infection occurring at a higher frequency than has been reported for P. aeruginosa (5, 37, 52). In addition, we found that even clonally related S. aureus lineages can change methicillin-resistance status within a patient over time, likely reflecting dynamic loss or gain of the resistance-determining SCCmec cassette (42). These findings underscore the limitations of current, isolate-focused microbiological practices (50) in characterizing S. aureus antibiotic susceptibility. Although the relationship between antimicrobial-susceptibility testing and therapeutic response in CF is unclear (53), our data suggest that approaches capable of identifying coexistent, phenotypically distinct populations could enable improved therapeutic selections accounting for this diversity. Culture-based or simplified molecular screens could feasibly be developed to affordably characterize these patterns identified by WGS.

In addition to high-resolution molecular epidemiology, WGS also revealed details of the genetic changes undergone by S. aureus during chronic CF respiratory infection. Although hypermutation is believed to contribute to S. aureus adaptation in CF (33), inferred mutation rates in this cohort were no greater than those reported during asymptomatic carriage in healthy populations (22). In addition, despite in vitro data suggesting that SCVs are hypermutable (54), in vivo mutation rates for SCV and normal-colony isolates were indistinguishable. S. aureus mutational adaptation in CF consequently appears to be driven by selection for spontaneous mutations in specific genes rather than by genome-wide hypermutation. We identified convergent evolution in several such genes, evidenced by their positive selection in multiple children, sometimes involving the same mutation. Their role in the establishment of chronic CF respiratory infection is likely explained by phenotypes related to bacterial persistence (agrA, rsbU, and ebh), antimicrobial resistance (rpsJ), or both (rpoB and thyA).

Downregulation of the agr quorum-sensing system (55) and sigB regulon (21) favor S. aureus phenotypic switching from acute to chronic states of infection. Genes agrA and rsbU, both strongly selected for loss of function, regulate these respective systems, and clinical isolates with corresponding de novo mutations empirically demonstrated attenuated virulence phenotypes. We previously showed that disruption of another positively selected gene, membrane protein ebh, increases intracellular invasion and persistence in vitro (56). Collectively, these results provide evidence that enhanced persistence phenotypes are common in S. aureus strains evolved to the CF airway and are enacted by mutation of key genes.

Selected rpsJ mutations were associated with tigecycline treatment and clustered near a protein loop known to mediate tigecycline resistance (57). This study newly uncovered that the same mutations were independently associated with fluoroquinolone therapy and reduced susceptibility to levofloxacin, demonstrating a broader role for this gene in the development of antimicrobial resistance in CF.

Of all genes, RNA polymerase rpoB showed the strongest signal for selection. Most mutations involved residue H481, known to confer resistance to rifamycin (58). However, selection for these mutations could only be partially accounted for by rifamycin exposure, and the mutations did not impact in vitro susceptibilities to other antibiotics, making it unlikely that rpoB mutations in CF are driven primarily by antimicrobial resistance. rpoB H481 has also been shown to reduce virulence and promote persistence through resistance to host antimicrobial peptides (59), providing a plausible explanation. Analogously, inactivation of thyA was associated with trimethoprim–sulfamethoxazole treatment and is a well-established cause of the persistent SCV phenotype seen in CF (39). The identification of two frequently mutated genes conferring dual antimicrobial-resistance and persistence phenotypes supports a paradigm in which adaptations favoring chronic infection are initially selected by antimicrobial therapies but do not revert after cessation of treatment and become “fixed” on the basis of their enhanced persistence phenotypes.

Our study has several limitations. Isolates may have adapted before our earliest collections, preventing identification of key genes subject to rapid selection. In addition, details regarding antibiotic usage were only available during the study period, requiring focus on mutations occurring after known exposures. Finally, bacteria were cultured from sputum and oropharyngeal swabs, as BAL is not commonly performed in children and may not reflect the microbiology of the lower airways.

Despite these limitations, our findings have important clinical implications. First, the frequent occurrence of persistence-associated rpoB and thyA mutations in S. aureus CF isolates and their association with rifampin and trimethoprim–sulfamethoxazole exposure suggest that although eradication regimens using these agents may achieve short-term culture clearance (16), they may paradoxically promote chronic infection with persistent strains that are difficult to identify and treat. This scenario underscores the importance of including long-term outcomes in clinical trials of MRSA-eradication regimens (10, 11, 16, 17), including surveillance for SCVs. Second, genes commonly altered in establishment of chronic respiratory infection identify alternative targets for S. aureus treatment in CF. Both agrA and alternative sigma factors regulate bacterial pathogenesis and were under strong positive selection in this study. Recently described agents targeting those pathways (60, 61) have the advantage of decoupling selection for bacterial persistence from antimicrobial resistance and could rationally be applied to early S. aureus infection in CF. Finally, physicians should consider the potential of patients with CF to harbor multiple S. aureus populations differing in antibiotic susceptibility when selecting therapies, and laboratory testing practices capable of assessing this heterogeneity must be studied. Collectively, our results demonstrate the promise of sequencing-based surveillance to provide comprehensive, individualized, and clinically relevant information for infection staging and treatment in CF or other chronic respiratory infections.